A Fun & Absurd Introduction to Vector Databases

A Fun & Absurd Introduction to Vector Databases

by Alexander Chatzizacharias

Formal & Physical Sciences Computing & Cybersecurity UNDatabasesUNHInformation retrieval
🎙 Alexander Chatzizacharias 👥 1.1M 📅 June 10, 2026 ⏱ 45 min 👁 3K 📄 tutorial 🧭 2026-08-02
Available in: English (current) Français

Keywords

vector databaseembeddingsemantic searchHNSWnearest neighbor

Summary

In this GOTO Copenhagen 2025 talk, Alexander Chatzizacharias provides an engaging and humorous introduction to vector databases. He begins by explaining vectors as mathematical constructs and their use in machine learning for representing semantic meaning. He covers the evolution of embedding models from Word2Vec to BERT and CLIP, highlighting the importance of context. The talk then defines vector databases as purpose-built systems for storing and querying vectors, emphasizing their role in semantic search. Chatzizacharias discusses indexing algorithms, particularly HNSW, and distance metrics like cosine similarity. He demonstrates the concepts through several playful demos, including a 3D visualization of a vector space and applications like searching for D&D items and Pokémon. The talk concludes with practical advice on getting started with vector databases and their potential beyond AI applications.

128 words

Critical Evaluation

The talk serves as an excellent introductory resource for developers unfamiliar with vector databases. Chatzizacharias employs a clear, accessible language and uses analogies and humor to demystify complex concepts. The structure is logical, progressing from basic vector definitions to practical implementations. The demos are particularly effective in illustrating how vector databases can be used for semantic search in various domains, from text to images. However, the presentation remains at a high level, lacking in-depth mathematical explanations or performance comparisons. The speaker does not cite specific academic papers or external sources, relying instead on his own experience and the conference context. The technical depth is moderate, suitable for a general technical audience but not for experts seeking advanced insights. The title’s promise of ‘fun and absurd’ is fulfilled through the playful demos and the speaker’s engaging style. Overall, the talk is a valuable primer that inspires further exploration, but it does not break new ground or offer rigorous scientific analysis.

159 words

Title / Content Match

The title accurately reflects the content: a fun and accessible introduction to vector databases, with a playful tone and interactive demos.

Quality & Reliability

7/10

The talk provides a clear and engaging introduction to vector databases, covering fundamental concepts such as vectors, embedding models, and indexing algorithms. The speaker demonstrates practical applications through demos, but the content is introductory and lacks deep technical detail or rigorous mathematical treatment. Sources are primarily the speaker's own resources and conference materials, with no direct citations to academic papers.

Chapters

Cited Sources

Concurring Sources

External References

Contribution & Novelties

The talk provides a playful and accessible introduction to vector databases, emphasizing their use beyond AI applications. It demystifies concepts like embeddings and HNSW through interactive demos, making the technology approachable for developers.

Pour aller plus loin :

  • Word2Vec paper — Original paper introducing Word2Vec, foundational to embedding models.
  • BERT paper — Introduces BERT, a transformer-based model that captures context in embeddings.
  • HNSW paper — Describes Hierarchical Navigable Small World graphs, a key indexing algorithm for vector databases.
  • CLIP paper — Introduces CLIP, a multimodal model that aligns images and text in a shared embedding space.

96 words

Radar Profile

The radar profile shows strong scores in information quantity and quality, with a moderate technical level. The talk is well-structured and informative, but the technical depth is limited, making it more suitable for beginners than experts.

Reliability 7/10